Restructure the shared memory allocation and fixed multiple bugs in the multicore...

Restructure the shared memory allocation and fixed multiple bugs in the multicore gc. Now the startegy of shared memory allocation can be configurable as 1) each core can only use its local memory 2) use local memory for lower address space while use global memory for higher address space or 3) all cores allocate globally. The simplest test case can now go through several round of gc correctly. Still need more tests. Also organized the codes a little bit.